Registry entries

The registry's list. Entries R-2026-001 and R-2026-002 are Cleanstone's own Tree A predictions, entered retroactively on 2026-10-30. They do not meet the protocol: there was one camp, the executor held the claim, and the freeze was a plan file in a working session, not a dated public post. They are listed so the record starts with what was actually run, labelled for what it is. R-2026-003 is the first entry meant to meet the protocol in full.

R-2026-001, retroactive

Entry id:            R-2026-001
Status:              executed (retroactive; protocol not met, see above)
Claim:               Delayed gratification and algotype clustering in
                     Zhang, Goldstein, Levin (2024, arXiv 2401.05375)
                     survive scheduler control and metric substitution.
Models:              classical and cell-view bubble, insertion, selection;
                     stdlib Python 3.11.2
Nulls:               zero frozen cells; label-only chimera; two schedulers
Primary metric:      delayed gratification, authors' formula and ours
Alternative metric:  total inversions (Kendall)
Sample size:         n = 64 and 100; 100 seeds per condition
Camp A:              Zhang, Goldstein, Levin; not invited before the run;
                     write-up sent 2026-10-30
Camp B:              Cleanstone
  Predictions:       P1 DG > 0 at zero barriers; P2 Kendall DG = 0 for
                     bubble and insertion; P3 DG rises with frozen cells,
                     small effect; P4 clustering tracks the efficiency gap;
                     P5 invariant to scheduler
Executor:            Cleanstone (holds Camp B; protocol not met)
Freeze:              ../../trees/01-a-ledger/plan.md, Experiment 0.1, written
                     before code on 2026-09-30; commit bd5b24d43599 holds plan,
                     code and results together, so order is attested by the
                     session transcript only (writeup.md section 2)

Result:              P1 held. P2 held at every setting. P3 held under the
                     authors' formula and reversed under ours. P4 held through
                     R-2026-002. P5 failed: bubble 0.218 vs 0.388 across
                     schedulers. ../../trees/01-a-ledger/writeup.md 3.3
Executed:            2026-09-30
Camp A reply:        awaited; follow-up due 2026-11-13

R-2026-002, retroactive

Entry id:            R-2026-002
Status:              executed (retroactive; protocol not met)
Claim:               Algotype clustering in chimeric arrays is produced by
                     differences in moves per round.
Models:              bubble rule only; half the cells labelled B and move
                     with probability p per turn
Nulls:               p = 1 (labels dynamically identical)
Primary metric:      mean same-label-neighbour fraction, bootstrap CI
Sample size:         n = 100, 100 seeds per p
Camp B:              Cleanstone
  Predictions:       P4a at p = 1 equals the label-only control (0.244);
                     P4b mean rises as p falls; P4c some p reaches the
                     bubble+insertion chimera (0.319)
Freeze:              ../../trees/01-a-ledger/plan.md, Experiment 0.1b,
                     before code, 2026-09-30

Result:              P4a held (0.244). P4b held. P4c not reached in the
                     pre-registered set (0.294 at p = 0.1); post hoc p = 0.02
                     and 0.01 reach 0.316 and 0.320. writeup.md 3.6
Executed:            2026-09-30